Where scale genuinely compounds
Three places. Fixed cost amortization, since peering, hardware and compliance are committed in advance and every additional gigabyte spreads them thinner. Demand diversity, because uncorrelated buyers fill the gaps each other leave, and the twentieth buyer improves fill more than the second improves price. And measurement, since a probe fleet that costs the same to run gets more statistically useful with every address it covers.
The second one is the interesting one, because it is not a purchasing advantage. It is a scheduling advantage, and it cannot be bought by a larger competitor without also acquiring the demand.
Buying more bandwidth cheaply is procurement. Filling the bandwidth you have is a different business.
Where it does not
Arbitration is linear. Every dispute costs roughly the same to settle whether you run a hundred addresses or a million, and disputes grow with volume. Consent auditing is linear for the same reason. Support is worse than linear if the product is confusing, which is a real argument for building a product that does not need explaining.
We plan on the assumption that these costs never fall per unit, which is why they are funded from a percentage rather than a subscription. If they scaled, we would have priced them into a flat fee and been wrong.
What that implies about the price floor
The list price can keep falling toward the marginal cost of moving traffic, but not below the linear cost of arbitrating and auditing it. That floor is somewhere around a cent and a half a gigabyte on current volumes, and it moves down slowly as automation replaces judgement in the easy disputes.
